A non-negotiable lower bound below which no efficiency argument, consensus decision, or governance outcome can push a legitimate collaboration design.
The HEXAD architecture (HC-026) creates a governance structure. The veto mechanism (HC-027) protects structural minorities within that structure. But governance structures can be captured. Vetos can be bargained away. Supermajorities can form around expedient decisions that degrade human capability. There must be a floor — a lower bound that no governance outcome, however broadly supported, can breach.
The Human Anchor Principle establishes that floor. It is derived from the Fidelity criterion: humans in a collaboration must not become less capable over time in the domain's irreducible functions. The Anchor takes the Fidelity criterion and makes it non-negotiable. Not subject to governance. Not subject to consensus. Not subject to efficiency arguments or competitive pressures or market logic. A sovereignty floor below which the collaboration design is illegitimate regardless of who approved it.
This paper operationalizes the Anchor across eight domains — specifying, for each, the minimum human capability requirements that constitute the floor.
Kant (1785) in the Groundwork of the Metaphysics of Morals established the principle that has never been improved upon: humanity must be treated as an end in itself, never merely as a means. The formulation is precise and its application to AI governance is direct. When an AI system treats human capability as a cost to be minimized — when efficiency gains are measured by how much human judgment can be replaced rather than how much human capability can be enhanced — humanity is being treated as means, not end.
The Universal Declaration of Human Rights (1948) operationalizes the Kantian principle in Articles 1 (inherent dignity and equal rights), 3 (right to life, liberty, and security of person), and 12 (protection from arbitrary interference). These are not aspirational statements. They are the documented minimum below which no state action — and by extension, no deployment of state-authorized or state-tolerated automated systems — is legitimate.
The Human Anchor translates these principles from philosophy and international law into domain-specific operational requirements. The floor is not abstract. It is concrete, measurable, and enforceable.
The question is not whether AI can perform the function. It is whether human capability to perform the function is something we are willing to lose.
The EU AI Act Article 5 prohibits five categories of AI practice: social scoring by governments, real-time remote biometric identification in public spaces (with exceptions), exploitation of vulnerabilities of specific groups, subliminal manipulation causing harm, and emotion recognition in workplaces and education (with exceptions). These prohibitions constitute the closest existing legal sovereignty floor.
The gaps are documented. Article 5 prohibits specific practices but does not establish a general principle. It does not prohibit AI deployments that systematically degrade human capability in domains not covered by its specific prohibitions. It does not address the slow erosion of professional judgment through AI-mediated workflow design. It does not protect the developmental experiences of children beyond the specific vulnerability exploitation clause. The EU AI Act is a floor with holes — better than no floor, but structurally insufficient as a sovereignty standard.
The Human Anchor fills the gaps. Not by listing prohibited practices (which will always lag behind technological capability) but by establishing a principle: in each domain, there are irreducible human capabilities that constitute the sovereignty floor. Any deployment that degrades those capabilities below the floor is illegitimate, regardless of its efficiency gains, regardless of its governance approval, regardless of its market success.
Children retain developmental experiences that cannot be AI-mediated without capability loss: relational learning (the formation of knowledge through human relationship), moral formation (the development of ethical judgment through human modeling and dialogue), and social-emotional development during critical periods (attachment, empathy, conflict resolution). AI tutoring that replaces these experiences — rather than supplementing the non-relational components of education — breaches the floor.
Patients retain the right to human clinical judgment at critical junctures: diagnosis communication (the moment when a human clinician translates clinical findings into meaning for the patient's life), treatment decisions requiring values integration (where the patient's values, not just their clinical data, determine the course of care), and end-of-life decisions (where the full weight of human presence, empathy, and moral accountability is irreducible). AI diagnostics that eliminate the human clinician from these junctures breach the floor.
Defendants retain the right to human decision-makers at sentencing and conviction. Chouldechova's impossibility theorem (2017) demonstrates that algorithmic risk assessment cannot simultaneously satisfy calibration, predictive parity, and balance — making algorithmic-only sentencing structurally unjust. The floor requires that a human judge, exercising human judgment with full moral accountability, makes the final determination in any proceeding that may deprive a person of liberty.
Citizens retain the right to human deliberation in collective decisions that bind them. Habermas's legitimacy condition requires that binding decisions emerge from deliberative processes in which all affected parties can participate as free and equal. AI systems that automate governance decisions — welfare eligibility, resource allocation, regulatory enforcement — without human deliberation at the point of decision breach the floor. The binding decision must be made by a human who can be held accountable for it.
Strategic financial judgment with moral accountability cannot be fully delegated. Algorithmic trading can execute. Portfolio optimization can calculate. But the strategic decisions that determine whose capital is deployed toward what ends — decisions with moral weight that affects communities, industries, and livelihoods — require human judgment that carries human accountability. The floor prohibits full delegation of strategic financial authority to automated systems.
Craft judgment and safety assessment require embodied human presence. A structural engineer must physically inspect a load-bearing element to exercise the judgment that protects human life. A master carpenter reads grain, tension, and material behavior through touch and experience that cannot be fully captured by sensors. The floor requires that safety-critical assessments in construction retain embodied human judgment — not because sensors are inaccurate, but because the consequences of error demand the full accountability of human presence.
Hypothesis formation and research ethics require human judgment. AI can process data, identify patterns, and generate candidate hypotheses. But the act of deciding which questions are worth asking — which hypotheses merit investigation, which research directions serve human knowledge rather than commercial optimization — requires human judgment shaped by curiosity, values, and moral reasoning. Research ethics review, in particular, cannot be automated: the assessment of whether a study design respects human dignity requires the exercise of human dignity.
Therapeutic presence cannot be substituted. The therapeutic alliance — the relationship between caregiver and care-receiver that produces healing — requires human presence: attunement, empathy, co-regulation, and the moral weight of one human bearing witness to another's suffering. AI systems that simulate therapeutic presence without providing it — chatbots marketed as mental health support, care robots presented as companions — breach the floor not because they fail to help, but because they substitute a simulation for the irreducible human act of being present with another person in their vulnerability.
Zuboff (2019) documented the most comprehensive sovereignty floor violation in current AI deployment: surveillance capitalism. The business model treats human behavioral data as raw material for prediction products — treating humanity as means (data source) rather than end (beneficiary). The violation is not incidental. It is the business model. The entire architecture of behavioral surplus extraction is designed to maximize the conversion of human experience into prediction revenue.
Under the Human Anchor Principle, surveillance capitalism as described by Zuboff is a structural floor violation across all eight domains. It degrades human autonomy (governance floor), instrumentalizes human relationships (care floor), commodifies developmental experiences (education floor), and undermines informed consent (healthcare floor). The violation is not that data is collected. It is that human experience is systematically converted into a means for someone else's end — the precise Kantian violation that the Anchor is designed to prevent.
"The Human Anchor is paternalistic. It prevents people from choosing to delegate their capabilities to AI if they find it beneficial. Who decides what the floor is?" The objection is serious and the answer is structural. The floor is not set by authority. It is derived from the Fidelity criterion, which is itself proposed based on capability analysis in each domain. The education floor draws on developmental psychology. The healthcare floor draws on clinical ethics. The law floor draws on mathematical impossibility theorems. The floors are proposed normative standards — they represent the Institute’s assessment of the minimum below which the domain ceases to serve human capability, and require domain expert review and empirical validation before regulatory application. Individuals may choose to use AI above the floor in any way they wish. The floor prevents systems — not individuals — from being designed to operate below it.
This paper completes Series 5: The HEXAD Applied. The four-paper arc has moved from diagnosis (HC-025, the governance gap) through architecture (HC-026, the six-node structure) through protection (HC-027, the veto mechanism) to principle (HC-028, the non-negotiable floor). Together, they constitute a governance framework for AI deployment that addresses the structural failures documented throughout The Collaboration.
The governance gap (HC-025) identified the problem: unrepresentative stakeholder governance. The HEXAD translation (HC-026) proposed the architecture: six nodes with supermajority, veto, and Human Anchor rules. The minority protection standard (HC-027) provided the safeguard: structural veto preventing majority-stakeholder harm. The Human Anchor Principle (HC-028) established the floor: the non-negotiable minimum below which no governance outcome can push.
The framework is complete. What remains is implementation — the subject of the broader Saga XI resolution.
Internal: This paper is part of The Collaboration (HC series), Saga XI. It draws on and contributes to the argument documented across 31 papers in 2 series.
External references for this paper are in development. The Institute’s reference program is adding formal academic citations across the corpus. Priority papers (P0/P1) have complete references sections.